Reinforcement Learning for Intraoperative Hypotension Management with Consideration to Postoperative Acute Kidney...

Area of Science:

• Anesthesiology
• Artificial Intelligence
• Nephrology

Background:

• Traditional surgical decision-making relies on variable human experience.
• Suboptimal management of intraoperative hypotension is linked to postoperative acute kidney injury (AKI).
• A data-driven system can aid perioperative decision-making for hypotension.

Purpose of the Study:

• To develop a deep reinforcement learning model for optimizing perioperative management.
• To recommend optimal doses of intravenous fluids and vasopressors to prevent intraoperative hypotension and subsequent AKI.

Main Methods:

• Retrospective analysis of 50,021 surgeries from 42,547 adult patients (2014-2020).
• Development of a deep reinforcement learning model to recommend fluid and vasopressor dosages.
• AKI definition based on Kidney Disease: Improving Global Outcomes (KDIGO) serum creatinine criteria.

Main Results:

• The model aligned with 69% of physician vasopressor decisions, with deviations in 10% (higher) and 21% (lower) of cases.
• For intravenous fluids, model recommendations matched actual doses within 0.05 ml/kg/15 min in 41% of cases.
• The reinforcement learning policy demonstrated a higher estimated value than physician or random policies, and AKI incidence was lowest with model-aligned dosages.

Conclusions:

• The developed AI model shows potential for reducing postoperative AKI.
• Implementing the model's policy may improve outcomes related to intraoperative hypotension.
• Data-driven recommendations can enhance perioperative care and patient safety.